Optimizing bottle-neck features for LVCSR

Bottle-neck, MLP structure, features, LVCSR

This publication deals with optimising various processing steps in Bottle-Neck feature extraction for lower word error rate on large vocabulary continuous speech recognition tasks.

This work continues in development of the recently proposed. Bottle-Neck features for ASR. A five-layers MLP used in bottle-neck  feature extraction allows to obtain arbitrary feature size without dimensionality reduction by transforms, independently on the MLP training targets. The MLP topology -- number and sizes of layers, suitable training targets, the impact of output feature transforms, the need of delta features, and the dimensionality of the final feature vector are studied with respect to the best ASR result. Optimized features are employed in three LVCSR tasks: Arabic broadcast news, English conversational telephone speech and English meetings. Improvements over standard cepstral features and probabilistic MLP features are shown for different tasks and different neural net input representations. A significant improvement is observed when phoneme MLP training targets are replaced by phoneme states and when delta features are added.